Description
What this service covers
A generic AI assistant can only guess about your business, because it has never seen your documentation. Our RAG service works over the material you supply, such as policies, manuals, help articles and contracts, stored as embeddings in a vector store, with retrieval tuned against a set of questions you have signed off. Every answer shows its source passage, and when your documents do not cover a question, the assistant says so. We measure accuracy and refusal behaviour rather than taking either on trust.
What you receive
- Ingestion of portable document files, word-processor files, web pages and wiki pages
- Embeddings and chunk sizes tuned to your documents, stored in a vector database
- Question-answering service that cites a source for every reply
- Chat widget for your website, or a bot inside your team chat
- 50-question evaluation set, with scores recorded pre- and post-tuning
- Safeguards against off-topic questions and prompt injection attempts
- Monthly cost projection based on your expected query volume
The work in 4 stages
-
01
Stage 1: Document preparation
Your documents are collected, duplicates removed and the content cleaned.
-
02
Stage 2: Assistant build
We build document loading, search and the reply generator.
-
03
Stage 3: Accuracy tuning
We score replies using your question set and adjust retrieval until results improve.
-
04
Stage 4: Pilot rollout
A pilot group gets access first, followed by wider release once accuracy holds steady.
Who it suits, and what we need from you
Ideal for
- Support teams answering the same policy questions every day
- Organisations with internal documents that no one can search
- Products that require users to read a long manual before starting
What we need from you
- Your document collection, with the latest version of each file marked
- 50 real user questions, each with its correct answer
- A choice of which LLM provider you are comfortable using
You share these through the brief on your order page after checkout.
What changes for your team
- Answers are grounded in your documents and cite a checkable source
- Frequently repeated questions no longer turn into support tickets
- A measured accuracy figure, not an impression from a demo
Questions about this service
Will it invent answers?
Is our data used to train the model?
What are the running costs?
This is a remote service: nothing is shipped physically, and the work is delivered into systems you control. See delivery and handover and refunds and cancellation for the full terms.
Reviews
There are no reviews yet.